Merge pull request #967 from parthenon-hpc-lab/jdolence/force_inline_inner

Force inline par_for_inner

Author: Yurlungur

CHANGELOG.md

@@ -36,6 +36,7 @@
 - [[PR 890]](https://github.com/parthenon-hpc-lab/parthenon/pull/890) Fix bugs in sparse communication and prolongation
 
 ### Infrastructure (changes irrelevant to downstream codes)
+- [[PR 967]](https://github.com/parthenon-hpc-lab/parthenon/pull/967) Change INLINE to FORCEINLINE on par_for_inner overloads
 - [[PR 938]](https://github.com/parthenon-hpc-lab/parthenon/pull/938) Restructure buffer packing/unpacking kernel hierarchical parallelism
 - [[PR 944]](https://github.com/parthenon-hpc-lab/parthenon/pull/944) Move sparse pack identifier creation to descriptor
 - [[PR 904]](https://github.com/parthenon-hpc-lab/parthenon/pull/904) Move to prolongation/restriction in one for AMR and communicate non-cell centered fields

src/kokkos_abstraction.hpp

@@ -1,9 +1,9 @@
 //========================================================================================
 // Parthenon performance portable AMR framework
-// Copyright(C) 2020-2022 The Parthenon collaboration
+// Copyright(C) 2020-2023 The Parthenon collaboration
 // Licensed under the 3-clause BSD License, see LICENSE file for details
 //========================================================================================
-// (C) (or copyright) 2020-2022. Triad National Security, LLC. All rights reserved.
+// (C) (or copyright) 2020-2023. Triad National Security, LLC. All rights reserved.
 //
 // This program was produced under U.S. Government contract 89233218CNA000001
 // for Los Alamos National Laboratory (LANL), which is operated by Triad
@@ -707,7 +707,7 @@ inline void par_for_outer(OuterLoopPatternTeams, const std::string &name,
 
 // Inner parallel loop using TeamThreadRange
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void
+KOKKOS_FORCEINLINE_FUNCTION void
 par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int ll, const int lu,
               const int ml, const int mu, const int nl, const int nu, const int kl,
               const int ku, const int jl, const int ju, const int il, const int iu,
@@ -741,7 +741,7 @@ par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int ll, const i
       });
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void
+KOKKOS_FORCEINLINE_FUNCTION void
 par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int ml, const int mu,
               const int nl, const int nu, const int kl, const int ku, const int jl,
               const int ju, const int il, const int iu, const Function &function) {
@@ -770,7 +770,7 @@ par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int ml, const i
                        });
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void
+KOKKOS_FORCEINLINE_FUNCTION void
 par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int nl, const int nu,
               const int kl, const int ku, const int jl, const int ju, const int il,
               const int iu, const Function &function) {
@@ -795,10 +795,10 @@ par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int nl, const i
                        });
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member,
-                                          const int kl, const int ku, const int jl,
-                                          const int ju, const int il, const int iu,
-                                          const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void
+par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int kl, const int ku,
+              const int jl, const int ju, const int il, const int iu,
+              const Function &function) {
   const int Nk = ku - kl + 1;
   const int Nj = ju - jl + 1;
   const int Ni = iu - il + 1;
@@ -815,9 +815,9 @@ KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternTTR, team_mbr_t team_m
   });
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member,
-                                          const int jl, const int ju, const int il,
-                                          const int iu, const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void
+par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member, const int jl, const int ju,
+              const int il, const int iu, const Function &function) {
   const int Nj = ju - jl + 1;
   const int Ni = iu - il + 1;
   const int NjNi = Nj * Ni;
@@ -828,22 +828,22 @@ KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternTTR, team_mbr_t team_m
   });
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternTTR, team_mbr_t team_member,
-                                          const int il, const int iu,
-                                          const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void par_for_inner(InnerLoopPatternTTR,
+                                               team_mbr_t team_member, const int il,
+                                               const int iu, const Function &function) {
   Kokkos::parallel_for(Kokkos::TeamThreadRange(team_member, il, iu + 1), function);
 }
 // Inner parallel loop using TeamVectorRange
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternTVR, team_mbr_t team_member,
-                                          const int il, const int iu,
-                                          const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void par_for_inner(InnerLoopPatternTVR,
+                                               team_mbr_t team_member, const int il,
+                                               const int iu, const Function &function) {
   Kokkos::parallel_for(Kokkos::TeamVectorRange(team_member, il, iu + 1), function);
 }
 
 // Inner parallel loop using FOR SIMD
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void
+KOKKOS_FORCEINLINE_FUNCTION void
 par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member, const int nl, const int nu,
               const int kl, const int ku, const int jl, const int ju, const int il,
               const int iu, const Function &function) {
@@ -859,10 +859,10 @@ par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member, const int nl, con
   }
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member,
-                                          const int kl, const int ku, const int jl,
-                                          const int ju, const int il, const int iu,
-                                          const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void
+par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member, const int kl, const int ku,
+              const int jl, const int ju, const int il, const int iu,
+              const Function &function) {
   for (int k = kl; k <= ku; ++k) {
     for (int j = jl; j <= ju; ++j) {
 #pragma omp simd
@@ -873,9 +873,9 @@ KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternSimdFor, team_mbr_t te
   }
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member,
-                                          const int jl, const int ju, const int il,
-                                          const int iu, const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void
+par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member, const int jl, const int ju,
+              const int il, const int iu, const Function &function) {
   for (int j = jl; j <= ju; ++j) {
 #pragma omp simd
     for (int i = il; i <= iu; i++) {
@@ -884,9 +884,9 @@ KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternSimdFor, team_mbr_t te
   }
 }
 template <typename Function>
-KOKKOS_INLINE_FUNCTION void par_for_inner(InnerLoopPatternSimdFor, team_mbr_t team_member,
-                                          const int il, const int iu,
-                                          const Function &function) {
+KOKKOS_FORCEINLINE_FUNCTION void par_for_inner(InnerLoopPatternSimdFor,
+                                               team_mbr_t team_member, const int il,
+                                               const int iu, const Function &function) {
 #pragma omp simd
   for (int i = il; i <= iu; i++) {
     function(i);